TL-PINN: Transfer Learning-Enhanced Physics-Informed Neural Network for Lithium-Ion Battery State of Charge Estimation

Abstract

Accurate state-of-charge (SOC) estimation is essential for the safe and efficient operation of battery management systems (BMS). To address the limited accuracy and poor reliability of conventional methods under complex operating conditions, a transfer learning-enhanced physics-informed neural network (TL-PINN) is proposed for SOC estimation. The proposed framework embeds the underlying dynamics of the equivalent circuit into the neural network through differential equation constraints. By jointly optimizing data reconstruction and physical constraints, electrochemical knowledge is explicitly incorporated into the network, enabling improved estimation accuracy and robustness under complex scenarios. Furthermore, maximum mean discrepancy (MMD)-based adaptation and fine-tuning strategies are introduced to facilitate efficient reuse of source-domain knowledge and enhance cross-domain generalization. Experiments conducted on both single-cell and real-world vehicle datasets demonstrate that the base PINN model achieves the best overall performance, with a root mean square error (RMSE) of 0.835%. In cross-temperature transfer scenarios, the RMSE remains below 3.0% across all target temperatures after fine-tuning. Moreover, after transferring the pretrained single-cell model to real-world vehicle operating data, the proposed method still achieves an average RMSE of 4.21%. These results demonstrate the superior adaptability and robustness of the proposed TL-PINN, highlighting its potential for reliable battery state monitoring in electric vehicles